Rv570 Z?

[DerekBaker]

http://www.beyond3d.com/resources/chip/119 says that the chip has 24 Z, i.e. 2 per pipeline.

A google search gave sites showing 1 Z per pipeline, i.e. 12.

Who's right?

Thanks

 

[Rys]

RV570 has a double Z-only rate.

 

[AnarchX]

RV570 has a double Z-only rate.

I thought it was only a rumor, because benchmarks do not show it:

[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              Radeon X1950 Series x86/MMX/3DNow!/SSE2 v2.0.6067 WinXP Release
Resolution          1024x768 @ 60.30Hz
Method              Flush

[b]Fillrate[/b]
--[b]32 bits[/b]---------------------------------------
  Mode              R8G8B8A8 Z24 S8
  Col               6.655 GPix/s
[B][SIZE="3"]  Z                 6.790 GPix/s[/SIZE][/B]
  ColZ              5.373 GPix/s
  ZPassColZ         4.345 GPix/s
  ZCullLEqual       94.014 GPix/s
  ZCullGEqual       94.588 GPix/s
  ZCullEqual        6.783 GPix/s
  S                 6.794 GPix/s
  SCull             6.793 GPix/s
----[b]stencil test passed[/b]-------------------------
    S               6.509 GPix/s
    ZFailS          6.509 GPix/s
------[b]z test passed (LEQUAL)[/b]--------------------
      S             6.508 GPix/s
      ZS            6.509 GPix/s
      Col           6.653 GPix/s
      ColZ          5.373 GPix/s
      ColS          4.337 GPix/s
      ColZS         4.347 GPix/s



--[b]16 bits[/b]---------------------------------------
  Mode              R5G6B5A0 Z16 S0
  Col               6.792 GPix/s
  Z                 6.791 GPix/s
  ColZ              6.571 GPix/s
  ZPassColZ         5.285 GPix/s
  ZCullLEqual       94.588 GPix/s
  ZCullGEqual       94.587 GPix/s
  ZCullEqual        6.783 GPix/s


[b]Bandwidth[/b]
Mode                R8G8B8A8 Z24 S8
--[b]available to buffer clears[/b]--------------------
  All               48.230 GB/s
  Color             25.758 GB/s
  ZAndStencil       229.331 GB/s
  Z                 172.022 GB/s
  Stencil           6.313 GB/s

Draw                47.239 GB/s
BurnedByRAMDAC      189.688 MB/s
Physical            47.428 GB/s

[b]Geometry[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Plain vertices[/b]--------------------------------
  Fan               232.180 MTris/s
  List              79.856 MTris/s
  Clip              7.916 MTris/s

--[b]Vertex shading speed[/b]--------------------------
  LightD1           71.814 MTris/s
  LightP1           47.807 MTris/s
  LightP8           21.932 MTris/s


[b]Texturing[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               6.201 GPix/s
    2               3.324 GPix/s
    3               2.238 GPix/s
    4               1.687 GPix/s
    5               1.354 GPix/s
    6               1.130 GPix/s
    7               970.680 MPix/s
    8               850.348 MPix/s

----[b]Trilinear filter[/b]----------------------------
    1               3.351 GPix/s
    2               1.703 GPix/s
    3               1.142 GPix/s
    4               858.982 MPix/s
    5               687.871 MPix/s
    6               574.128 MPix/s
    7               492.550 MPix/s
    8               431.248 MPix/s



[b]Readback[/b]
Mode                R8G8B8A8 Z24 S8
--[b]Whole buffer[/b]----------------------------------
  R8G8B8A8          60.551 MPix/s
  B8G8R8A8          60.768 MPix/s
  R8G8B8            62.492 MPix/s
  B8G8R8            62.590 MPix/s
  Zuint             61.332 MPix/s
  Zfloat            59.883 MPix/s
  S8                1.666 MPix/s

--[b]32x32 region[/b]----------------------------------
  R8G8B8A8          16.856 MPix/s
  B8G8R8A8          17.008 MPix/s
  R8G8B8            17.127 MPix/s
  B8G8R8            16.975 MPix/s
  Zuint             15.137 MPix/s
  Zfloat            15.329 MPix/s
  S8                1.571 MPix/s


[b]Texture cache[/b]
Mode                R8G8B8A8 Z24 S8
RGBA                8 kiB
DXT1                8 kiB
DXT5                8 kiB

[b]Tiling[/b]
Mode                R8G8B8A8 Z24 S8
--[b]preferred block alignment[/b]---------------------
----[b]updating all buffers[/b]------------------------
    Width           4
    Height          4

----[b]in color buffer[/b]-----------------------------
    Width           4
    Height          4

----[b]in depth buffer[/b]-----------------------------
    Width           2
    Height          2

----[b]in stencil buffer[/b]---------------------------
    Width           2
    Height          2

On RV530 this benchmark measures the 8 Z:

[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              RADEON X1600 Series x86/MMX/3DNow!/SSE2 v2.0.5646 WinXP Release
Resolution          1024x768 @ 60.25Hz
Method              Flush

[b]Fillrate[/b]
--[b]32 bits[/b]---------------------------------------
  Mode              R8G8B8A8 Z24 S8
  Col               1.862 GPix/s
[B][SIZE="3"]  Z                 3.767 GPix/s[/SIZE][/B]
  ColZ              1.425 GPix/s
  ZPassColZ         1.133 GPix/s
  ZCullLEqual       27.946 GPix/s
  ZCullGEqual       27.943 GPix/s
  ZCullEqual        3.913 GPix/s
  S                 3.923 GPix/s
  SCull             3.923 GPix/s
----[b]stencil test passed[/b]-------------------------
    S               2.290 GPix/s
    ZFailS          2.290 GPix/s
------[b]z test passed (LEQUAL)[/b]--------------------
      S             2.290 GPix/s
      ZS            2.290 GPix/s
      Col           1.862 GPix/s
      ColZ          1.425 GPix/s
      ColS          1.134 GPix/s
      ColZS         1.134 GPix/s



--[b]16 bits[/b]---------------------------------------
  Mode              R5G6B5A0 Z16 S0
  Col               1.863 GPix/s
  Z                 3.463 GPix/s
  ColZ              1.787 GPix/s
  ZPassColZ         1.344 GPix/s
  ZCullLEqual       27.942 GPix/s
  ZCullGEqual       27.943 GPix/s
  ZCullEqual        3.913 GPix/s


[b]Bandwidth[/b]
Mode                R8G8B8A8 Z24 S8
--[b]available to buffer clears[/b]--------------------
  All               14.128 GB/s
  Color             7.225 GB/s
  ZAndStencil       121.866 GB/s
  Z                 91.396 GB/s
  Stencil           2.266 GB/s

Draw                12.930 GB/s
BurnedByRAMDAC      189.540 MB/s
Physical            13.119 GB/s

[b]Geometry[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Plain vertices[/b]--------------------------------
  Fan               197.378 MTris/s
  List              68.027 MTris/s
  Clip              6.888 MTris/s

--[b]Vertex shading speed[/b]--------------------------
  LightD1           59.425 MTris/s
  LightP1           26.184 MTris/s
  LightP8           12.005 MTris/s


[b]Texturing[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               1.796 GPix/s
    2               956.454 MPix/s
    3               653.422 MPix/s
    4               479.531 MPix/s
    5               394.249 MPix/s
    6               329.118 MPix/s
    7               282.368 MPix/s
    8               247.241 MPix/s

----[b]Trilinear filter[/b]----------------------------
    1               983.000 MPix/s
    2               497.194 MPix/s
    3               332.684 MPix/s
    4               250.028 MPix/s
    5               200.240 MPix/s
    6               167.004 MPix/s
    7               143.227 MPix/s
    8               125.377 MPix/s



[b]Readback[/b]
Mode                R8G8B8A8 Z24 S8
--[b]Whole buffer[/b]----------------------------------
  R8G8B8A8          52.637 MPix/s
  B8G8R8A8          57.646 MPix/s
  R8G8B8            60.107 MPix/s
  B8G8R8            60.124 MPix/s
  Zuint             57.394 MPix/s
  Zfloat            56.575 MPix/s
  S8                2.018 MPix/s

--[b]32x32 region[/b]----------------------------------
  R8G8B8A8          14.508 MPix/s
  B8G8R8A8          14.444 MPix/s
  R8G8B8            14.668 MPix/s
  B8G8R8            14.580 MPix/s
  Zuint             13.330 MPix/s
  Zfloat            13.276 MPix/s
  S8                1.832 MPix/s


[b]Texture cache[/b]
Mode                R8G8B8A8 Z24 S8
RGBA                8 kiB
DXT1                8 kiB
DXT5                8 kiB

[b]Tiling[/b]
Mode                R8G8B8A8 Z24 S8
--[b]preferred block alignment[/b]---------------------
----[b]updating all buffers[/b]------------------------
    Width           4
    Height          8

----[b]in color buffer[/b]-----------------------------
    Width           4
    Height          8

----[b]in depth buffer[/b]-----------------------------
    Width           2
    Height          2

----[b]in stencil buffer[/b]---------------------------
    Width           2
    Height          2

 

[CarstenS]

RV570 has a double Z-only rate.

AFAIK only with MSAA applied.

 

[Dave Baumann]

RV570 most closely follows R580 in terms of how it is derived.

 

[ShaidarHaran]

And RV570's little brother (RV560) produces the following results:
Stock 575/690 GPU/mem clocks:

[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              Radeon X1650 Series x86/SSE2 v2.0.6747 WinXP Release
Resolution          1024x768 @ unknown refresh rate
Method              Flush

[b]Fillrate[/b]
--[b]32 bits[/b]---------------------------------------
  Mode              R8G8B8A8 Z24 S8
  Col               3.497 GPix/s
  Z                 4.540 GPix/s
  ColZ              2.702 GPix/s
  ZPassColZ         2.046 GPix/s
  ZCullLEqual       65.781 GPix/s
  ZCullGEqual       65.849 GPix/s
  ZCullEqual        4.536 GPix/s
  S                 4.528 GPix/s
  SCull             4.531 GPix/s
----[b]stencil test passed[/b]-------------------------
    S               4.116 GPix/s
    ZFailS          4.115 GPix/s
------[b]z test passed (LEQUAL)[/b]--------------------
      S             4.115 GPix/s
      ZS            4.115 GPix/s
      Col           3.497 GPix/s
      ColZ          2.703 GPix/s
      ColS          2.045 GPix/s
      ColZS         2.044 GPix/s



--[b]16 bits[/b]---------------------------------------
  Mode              R5G6B5A0 Z16 S0
  Col               4.538 GPix/s
  Z                 4.540 GPix/s
  ColZ              3.935 GPix/s
  ZPassColZ         2.898 GPix/s
  ZCullLEqual       64.225 GPix/s
  ZCullGEqual       65.838 GPix/s
  ZCullEqual        4.536 GPix/s


[b]Bandwidth[/b]
Mode                R8G8B8A8 Z24 S8
--[b]available to buffer clears[/b]--------------------
  All               25.912 GB/s
  Color             13.746 GB/s
  ZAndStencil       147.765 GB/s
  Z                 108.113 GB/s
  Stencil           4.023 GB/s

Draw                23.157 GB/s
BurnedByRAMDAC      267.387 MB/s
Physical            23.424 GB/s

[b]Geometry[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Plain vertices[/b]--------------------------------
  Fan               231.847 MTris/s
  List              79.716 MTris/s
  Clip              7.915 MTris/s

--[b]Vertex shading speed[/b]--------------------------
  LightD1           71.964 MTris/s
  LightP1           48.452 MTris/s
  LightP8           21.955 MTris/s


[b]Texturing[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               4.260 GPix/s
    2               2.233 GPix/s
    3               1.498 GPix/s
    4               1.128 GPix/s
    5               904.083 MPix/s
    6               753.495 MPix/s
    7               647.235 MPix/s
    8               566.594 MPix/s

----[b]Trilinear filter[/b]----------------------------
    1               2.207 GPix/s
    2               1.141 GPix/s
    3               763.671 MPix/s
    4               574.071 MPix/s
    5               459.741 MPix/s
    6               383.425 MPix/s
    7               328.758 MPix/s
    8               287.791 MPix/s



[b]Readback[/b]
Mode                R8G8B8A8 Z24 S8
--[b]Whole buffer[/b]----------------------------------
  R8G8B8A8          35.634 MPix/s
  B8G8R8A8          40.814 MPix/s
  R8G8B8            41.517 MPix/s
  B8G8R8            41.424 MPix/s
  Zuint             41.204 MPix/s
  Zfloat            41.059 MPix/s
  S8                2.073 MPix/s

--[b]32x32 region[/b]----------------------------------
  R8G8B8A8          9.735 MPix/s
  B8G8R8A8          9.622 MPix/s
  R8G8B8            9.851 MPix/s
  B8G8R8            9.826 MPix/s
  Zuint             8.299 MPix/s
  Zfloat            8.155 MPix/s
  S8                1.777 MPix/s


[b]Texture cache[/b]
Mode                R8G8B8A8 Z24 S8
RGBA                8 kiB
DXT1                8 kiB
DXT5                8 kiB

[b]Tiling[/b]
Mode                R8G8B8A8 Z24 S8
--[b]preferred block alignment[/b]---------------------
----[b]updating all buffers[/b]------------------------
    Width           4
    Height          4

----[b]in color buffer[/b]-----------------------------
    Width           4
    Height          4

----[b]in depth buffer[/b]-----------------------------
    Width           2
    Height          2

----[b]in stencil buffer[/b]---------------------------
    Width           2
    Height          2

Z is higher, but not double. Odd... AFAIK RV560 is just a cut-down RV570, i.e. 80nm RV570 with one quad of rops/tmus/alus disabled and half the memory controller bit-width. RGBA8 mode produces Z-only rates ~33% higher, but this gain is nullified in 5-6-5-0 mode

Let's see how she fares with higher mem clocks.
575/790 (max. stable mem clock):

[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              Radeon X1650 Series x86/SSE2 v2.0.6747 WinXP Release
Resolution          1024x768 @ unknown refresh rate
Method              Flush

[b]Fillrate[/b]
--[b]32 bits[/b]---------------------------------------
  Mode              R8G8B8A8 Z24 S8
  Col               3.526 GPix/s
  Z                 4.541 GPix/s
  ColZ              3.029 GPix/s
  ZPassColZ         2.298 GPix/s
  ZCullLEqual       65.829 GPix/s
  ZCullGEqual       65.854 GPix/s
  ZCullEqual        4.535 GPix/s
  S                 4.541 GPix/s
  SCull             4.541 GPix/s
----[b]stencil test passed[/b]-------------------------
    S               4.411 GPix/s
    ZFailS          4.413 GPix/s
------[b]z test passed (LEQUAL)[/b]--------------------
      S             4.414 GPix/s
      ZS            4.408 GPix/s
      Col           3.528 GPix/s
      ColZ          3.026 GPix/s
      ColS          2.302 GPix/s
      ColZS         2.302 GPix/s



--[b]16 bits[/b]---------------------------------------
  Mode              R5G6B5A0 Z16 S0
  Col               4.532 GPix/s
  Z                 4.540 GPix/s
  ColZ              4.249 GPix/s
  ZPassColZ         3.304 GPix/s
  ZCullLEqual       65.749 GPix/s
  ZCullGEqual       65.848 GPix/s
  ZCullEqual        4.536 GPix/s


[b]Bandwidth[/b]
Mode                R8G8B8A8 Z24 S8
--[b]available to buffer clears[/b]--------------------
  All               26.070 GB/s
  Color             13.840 GB/s
  ZAndStencil       144.191 GB/s
  Z                 108.514 GB/s
  Stencil           4.326 GB/s

Draw                26.305 GB/s
BurnedByRAMDAC      267.387 MB/s
Physical            26.573 GB/s

[b]Geometry[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Plain vertices[/b]--------------------------------
  Fan               231.922 MTris/s
  List              79.748 MTris/s
  Clip              7.916 MTris/s

--[b]Vertex shading speed[/b]--------------------------
  LightD1           71.995 MTris/s
  LightP1           47.811 MTris/s
  LightP8           21.952 MTris/s


[b]Texturing[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               4.353 GPix/s
    2               2.230 GPix/s
    3               1.496 GPix/s
    4               1.126 GPix/s
    5               899.919 MPix/s
    6               753.390 MPix/s
    7               645.687 MPix/s
    8               565.510 MPix/s

----[b]Trilinear filter[/b]----------------------------
    1               2.253 GPix/s
    2               1.092 GPix/s
    3               747.256 MPix/s
    4               573.006 MPix/s
    5               458.852 MPix/s
    6               375.812 MPix/s
    7               328.242 MPix/s
    8               287.206 MPix/s



[b]Readback[/b]
Mode                R8G8B8A8 Z24 S8
--[b]Whole buffer[/b]----------------------------------
  R8G8B8A8          35.146 MPix/s
  B8G8R8A8          40.470 MPix/s
  R8G8B8            41.372 MPix/s
  B8G8R8            41.411 MPix/s
  Zuint             41.117 MPix/s
  Zfloat            40.523 MPix/s
  S8                2.076 MPix/s

--[b]32x32 region[/b]----------------------------------
  R8G8B8A8          9.589 MPix/s
  B8G8R8A8          9.730 MPix/s
  R8G8B8            9.776 MPix/s
  B8G8R8            9.807 MPix/s
  Zuint             8.200 MPix/s
  Zfloat            8.270 MPix/s
  S8                1.780 MPix/s


[b]Texture cache[/b]
Mode                R8G8B8A8 Z24 S8
RGBA                8 kiB
DXT1                8 kiB
DXT5                8 kiB

[b]Tiling[/b]
Mode                R8G8B8A8 Z24 S8
--[b]preferred block alignment[/b]---------------------
----[b]updating all buffers[/b]------------------------
    Width           4
    Height          4

----[b]in color buffer[/b]-----------------------------
    Width           4
    Height          4

----[b]in depth buffer[/b]-----------------------------
    Width           2
    Height          2

----[b]in stencil buffer[/b]---------------------------
    Width           2
    Height          2

Z rate is approximately the same, but stencil rate increased with the bandwidth increase.

How about higher GPU clock.
641/690 (max. stable GPU clock):

[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              Radeon X1650 Series x86/SSE2 v2.0.6747 WinXP Release
Resolution          1024x768 @ unknown refresh rate
Method              Flush

[b]Fillrate[/b]
--[b]32 bits[/b]---------------------------------------
  Mode              R8G8B8A8 Z24 S8
  Col               3.868 GPix/s
  Z                 5.073 GPix/s
  ColZ              2.780 GPix/s
  ZPassColZ         2.103 GPix/s
  ZCullLEqual       73.386 GPix/s
  ZCullGEqual       73.358 GPix/s
  ZCullEqual        5.069 GPix/s
  S                 5.076 GPix/s
  SCull             5.072 GPix/s
----[b]stencil test passed[/b]-------------------------
    S               4.244 GPix/s
    ZFailS          4.246 GPix/s
------[b]z test passed (LEQUAL)[/b]--------------------
      S             4.246 GPix/s
      ZS            4.246 GPix/s
      Col           3.871 GPix/s
      ColZ          2.779 GPix/s
      ColS          2.104 GPix/s
      ColZS         2.105 GPix/s



--[b]16 bits[/b]---------------------------------------
  Mode              R5G6B5A0 Z16 S0
  Col               5.073 GPix/s
  Z                 5.065 GPix/s
  ColZ              4.069 GPix/s
  ZPassColZ         2.963 GPix/s
  ZCullLEqual       73.578 GPix/s
  ZCullGEqual       73.596 GPix/s
  ZCullEqual        5.070 GPix/s


[b]Bandwidth[/b]
Mode                R8G8B8A8 Z24 S8
--[b]available to buffer clears[/b]--------------------
  All               28.797 GB/s
  Color             15.270 GB/s
  ZAndStencil       145.060 GB/s
  Z                 99.208 GB/s
  Stencil           4.193 GB/s

Draw                24.021 GB/s
BurnedByRAMDAC      267.387 MB/s
Physical            24.288 GB/s

[b]Geometry[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Plain vertices[/b]--------------------------------
  Fan               258.475 MTris/s
  List              89.039 MTris/s
  Clip              8.846 MTris/s

--[b]Vertex shading speed[/b]--------------------------
  LightD1           80.337 MTris/s
  LightP1           54.076 MTris/s
  LightP8           24.525 MTris/s


[b]Texturing[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               4.862 GPix/s
    2               2.493 GPix/s
    3               1.674 GPix/s
    4               1.260 GPix/s
    5               1.009 GPix/s
    6               842.925 MPix/s
    7               723.534 MPix/s
    8               632.896 MPix/s

----[b]Trilinear filter[/b]----------------------------
    1               2.519 GPix/s
    2               1.275 GPix/s
    3               853.035 MPix/s
    4               640.893 MPix/s
    5               513.201 MPix/s
    6               428.237 MPix/s
    7               367.247 MPix/s
    8               321.639 MPix/s



[b]Readback[/b]
Mode                R8G8B8A8 Z24 S8
--[b]Whole buffer[/b]----------------------------------
  R8G8B8A8          35.850 MPix/s
  B8G8R8A8          40.871 MPix/s
  R8G8B8            41.048 MPix/s
  B8G8R8            41.476 MPix/s
  Zuint             41.683 MPix/s
  Zfloat            40.955 MPix/s
  S8                2.070 MPix/s

--[b]32x32 region[/b]----------------------------------
  R8G8B8A8          9.770 MPix/s
  B8G8R8A8          9.737 MPix/s
  R8G8B8            9.947 MPix/s
  B8G8R8            9.701 MPix/s
  Zuint             8.356 MPix/s
  Zfloat            8.374 MPix/s
  S8                1.786 MPix/s


[b]Texture cache[/b]
Mode                R8G8B8A8 Z24 S8
RGBA                8 kiB
DXT1                8 kiB
DXT5                8 kiB

[b]Tiling[/b]
Mode                R8G8B8A8 Z24 S8
--[b]preferred block alignment[/b]---------------------
----[b]updating all buffers[/b]------------------------
    Width           4
    Height          4

----[b]in color buffer[/b]-----------------------------
    Width           4
    Height          4

----[b]in depth buffer[/b]-----------------------------
    Width           2
    Height          2

----[b]in stencil buffer[/b]---------------------------
    Width           2
    Height          2

And now for combined max. clocks
635/790:

[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              Radeon X1650 Series x86/SSE2 v2.0.6747 WinXP Release
Resolution          1024x768 @ unknown refresh rate
Method              Flush

[b]Fillrate[/b]
--[b]32 bits[/b]---------------------------------------
  Mode              R8G8B8A8 Z24 S8
  Col               3.868 GPix/s
  Z                 5.021 GPix/s
  ColZ              3.074 GPix/s
  ZPassColZ         2.326 GPix/s
  ZCullLEqual       72.793 GPix/s
  ZCullGEqual       72.825 GPix/s
  ZCullEqual        5.017 GPix/s
  S                 5.020 GPix/s
  SCull             5.022 GPix/s
----[b]stencil test passed[/b]-------------------------
    S               4.654 GPix/s
    ZFailS          4.654 GPix/s
------[b]z test passed (LEQUAL)[/b]--------------------
      S             4.655 GPix/s
      ZS            4.656 GPix/s
      Col           3.874 GPix/s
      ColZ          3.070 GPix/s
      ColS          2.326 GPix/s
      ColZS         2.327 GPix/s



--[b]16 bits[/b]---------------------------------------
  Mode              R5G6B5A0 Z16 S0
  Col               5.002 GPix/s
  Z                 5.021 GPix/s
  ColZ              4.452 GPix/s
  ZPassColZ         3.327 GPix/s
  ZCullLEqual       72.807 GPix/s
  ZCullGEqual       72.822 GPix/s
  ZCullEqual        5.016 GPix/s


[b]Bandwidth[/b]
Mode                R8G8B8A8 Z24 S8
--[b]available to buffer clears[/b]--------------------
  All               28.653 GB/s
  Color             11.427 GB/s
  ZAndStencil       17.365 GB/s
  Z                 9.066 GB/s
  Stencil           3.129 GB/s

Draw                26.691 GB/s
BurnedByRAMDAC      267.387 MB/s
Physical            26.958 GB/s

[b]Geometry[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Plain vertices[/b]--------------------------------
  Fan               253.052 MTris/s
  List              88.155 MTris/s
  Clip              8.753 MTris/s

--[b]Vertex shading speed[/b]--------------------------
  LightD1           79.549 MTris/s
  LightP1           53.578 MTris/s
  LightP8           24.279 MTris/s


[b]Texturing[/b]
Mode                R5G6B5A0 Z16 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               4.810 GPix/s
    2               2.467 GPix/s
    3               1.657 GPix/s
    4               1.247 GPix/s
    5               999.455 MPix/s
    6               834.307 MPix/s
    7               715.696 MPix/s
    8               626.725 MPix/s

----[b]Trilinear filter[/b]----------------------------
    1               2.494 GPix/s
    2               1.261 GPix/s
    3               844.531 MPix/s
    4               634.450 MPix/s
    5               508.334 MPix/s
    6               423.919 MPix/s
    7               363.588 MPix/s
    8               318.275 MPix/s



[b]Readback[/b]
Mode                R8G8B8A8 Z24 S8
--[b]Whole buffer[/b]----------------------------------
  R8G8B8A8          35.390 MPix/s
  B8G8R8A8          40.844 MPix/s
  R8G8B8            41.499 MPix/s
  B8G8R8            41.602 MPix/s
  Zuint             41.345 MPix/s
  Zfloat            40.937 MPix/s
  S8                2.076 MPix/s

--[b]32x32 region[/b]----------------------------------
  R8G8B8A8          9.853 MPix/s
  B8G8R8A8          9.795 MPix/s
  R8G8B8            9.909 MPix/s
  B8G8R8            9.969 MPix/s
  Zuint             8.363 MPix/s
  Zfloat            8.333 MPix/s
  S8                1.785 MPix/s


[b]Texture cache[/b]
Mode                R8G8B8A8 Z24 S8
RGBA                8 kiB
DXT1                8 kiB
DXT5                8 kiB

[b]Tiling[/b]
Mode                R8G8B8A8 Z24 S8
--[b]preferred block alignment[/b]---------------------
----[b]updating all buffers[/b]------------------------
    Width           4
    Height          4

----[b]in color buffer[/b]-----------------------------
    Width           4
    Height          4

----[b]in depth buffer[/b]-----------------------------
    Width           2
    Height          2

----[b]in stencil buffer[/b]---------------------------
    Width           2
    Height          2

Ok, obviously rates are almost entirely GPU-bound, except stencil which appears to be bandwidth-limited.

Any thoughts on why the increase in Z rate in RGBA8 mode compared to 16-bit 5-6-5-0 mode?

 

[OpenGL guy]

Any thoughts on why the increase in Z rate in RGBA8 mode compared to 16-bit 5-6-5-0 mode?

There is no increased rate, there's a bandwidth limitation affecting your RGBA8 color results. When you go to 16-bit color, the bandwidth constraint is lifted and color rate can match Z.

I don't know why you didn't see much gain by increasing memory bandwidth, I'd've expected the color rate to improve some.